feipoa wrote on 2021-08-06, 00:13:

jakethompson1 wrote on 2021-08-05, 23:46:

Chadti99 wrote on 2021-08-05, 23:42:

Would these 12ns sram chips work on socket 3 boards for 256k cache?

https://www.ebay.com/itm/QTY-15-IS61LV256-12N … 1-127632-2357-0

Not sure what "low voltage" means but doesn't sound right. Should check the datasheet and see if they are 5 volts.

However, you can still get brand new 12ns 32KB chips. Like #71256SA12TPG at Digi-Key.

Have you ordered these? Could you share a photo? I recall someone here ordered them and wasn't impressed.

Yes, I did. Here was the thread with the picture: More 486 L2 cache troubles - timing sometimes so slow that cache isn't worth it?
I believe there is just a quirk with having the UM8672 VLB IDE controller on the bus and also using cache speeds any faster than 3-2-2-2. I have run into that issue with two different boards now. I don't think the Digi-Key chips were to blame.
In fact I'm using two of those chips as tag RAM chips on other boards and they work fine.

pshipkov, thank you. Those scores are interesting. Would you be willing to put the m918 scores next to the UUD scores, at least for 3x60 and 4x50?

I haven't had any issues with the IBM 5x86c at 2x66, however simple changes on the fringe, like different CPU or another multiplier can trigger instability.

I am surprised you can run 2-1-1-1 at 3x66. Is that really stable? The speedsys screen is exited before I can see if it is 256K or 512K. Which is it? [EDIT: 3x66 omits the speedsys screen, but 4x50 Mhz shows 256K. I assume 256K for all?] Did you try the UMC driver with Preempt PCI Master Option Disabled? I recall having some issues with this enabled for certain hardware, but I forget which now.

You should be able to get away with 1 WS at 66 MHz for DRAM read, but I noticed the board is very selective with the modules in this case. I'm surprised you need 8 BCLKs for I/O recovery time. Maybe the internal CPU speed is influencing this to an extent. I didn't need 8 BCLK at 2x66 with an X5 at 133 Mhz.

jakethompson1, thank you. I might try some of these, but the cost is so darn high. Another uses suspects they are fake, but maybe that's just because of the font colour and missing identifiers on the bottom surface? Perhaps this is the norm these days for all ICs? What happened to the white font anyway? Who thought it would be a good idea to go with light tan font that is nearly impossible to read?

M918i does not do 60MHz. It has some obscure clock generator. Was not able to find datasheet for it. Checked very carefully all possible options, but it looks like it maxes at 50.
Also, UUD is the better place by FAR, compared to i918i. 😀

---

3x66 is table with this (UUD) motherboard, given some buffering with relaxed BIOS timings.

The 3x66 run early today was a quick one. I am sure some things can be improved upon. That's for later.
L2 cache - 256Kb interleaved, 2 banks. Same setup as 4x50. I should keep SpeedSys to complete the memory test, but thought the video will get too long and boring.
2-1-1-1 is stable. Takes real good L2 chips.
As we talked before - this board is suuuper fussy about L2 cache. 😁

The 64Mb 50ns MB memory sticks are solid, but at 3x66 anything under 3 WS READ hangs. This is inconclusive of course. Will find out later.

UMC IDE - i think i tried that before. It didn't work.

Probably left the 8 BCLK by accident. It is fine at 2.

Heh, well what do you attribute the 1 fps improvement to with the m918 (21.8 fps) over the UUD board (20.8 fps)? Are you able to provide side-by-side speedsys screenshots?

Well, so far all PC-Chips M-something-something motherboards i touched are ... (was about to say complete garbage, but will say instead) ... not good.
They have some nice features of course and you can take them on a narrow vertical journey here and there, but if you zoom out and look at the whole package - janky.
I guess that's why there is so much chatter about them online - people trying to figure out what is going on.

Biostar UUD is not without its shortcomings - lacking IDE and L2 cache, but the overall design and integration of all its components is much better.
The performance bar considering complete system stability is way beyond any PC-Chips M### motherboard i know of.

I was thinking to post some details about M918i in the other thread tomorrow - that will clarify my stance.

Now now, PCChips motherboards are a Ferrari... with a corvette engine... and a porsche gearbox... and lambo suspension... just 'coz it doesn't all work together out of the box doesn't mean there's not potential 🤣

Heh. Thats another way of saying it.

That was a bit too much of a compliment if you ask me. More like:

a Ferrari with a corvette engine... with incorrect motor mounts, and a porshe gearbox... with only one mounting bolt instead of 3... and a lambo suspension... that bottoms out while going over bumps.

Getting warmer ...

For anyone interested, I got the following results at 180MHz(60x3) in GLQuake. All tests with sound @512x384 unless noted. Both boards with 256k cache and the tightest timings I could get.

LS486E-C2 w/3dfx Banshee = 23.3fps
LS486E-C2 w/3dfx Voodoo 2 = 24.5fps
PCI 1:1

8433UUD w/3dfx Banshee = no post
8433UUD w/3dfx Voodoo 2 = 26.6fps
8433UUD w/3dfx Voodoo 2 = 29.3fps* no sound
8433UUD w/3dfx Voodoo 2 = 29.1fps* no sound 640x480
PCI 1:2/3

Both the Banshee and Voodoo 2 seem happy at 60MHz PCI speeds.

The 8433 isn’t quite stable at these settings but I’m hoping to find the right combo soon.

So far this is the fastest I have seen. ( 200mhz )
https://youtu.be/LI1_RlVLhu8

Intel486dx33 wrote on 2021-08-06, 12:09:

So far this is the fastest I have seen. ( 200mhz )
https://youtu.be/LI1_RlVLhu8

Yes this is a great reference, I should rerun the tests without sound, it makes a difference.

Just for curiosity, can anyone try the latest build of 486quake (https://github.com/goshhhy/486quake/releases/tag/1.09-r6) with any of those 200MHz 486s?

ViTi95 wrote on 2021-08-06, 12:45:

Just for curiosity, can anyone try the latest build of 486quake (https://github.com/goshhhy/486quake/releases/tag/1.09-r6) with any of those 200MHz 486s?

AMD5x86@180(60x3) & Riva128
8433UUD w/256k cache
-nosound (same settings as Phil’s dosbench)

Quake 1.06 = 20.1
Quake 1.08 = 21
486Quake = 22.7
586Quake = 22

Very cool! Anything for more frames on this platform.

Still searching for the right memory timings on my 8433UUD v3.1 to get a “reliable” 180MHz. I’ve got double banked UMC 12ns sram installed. Tempted to swap in 512k. But maybe also I should try some thermal paste instead of just resting the cooler on top 🤣. Really enjoying messing with stuff, loving all the stats folks are sharing.

I was able to install Windows 98 on the LS486e-c2 and run GLQuake, on the Banshee, for a couple of hours no issues.

Was just really hoping to do the same on the 8433UUD and eek out just a hair more performance.

I know you are tweaking that for the sake of the sport, but not sure you will find greener grass there.
SIS is clock-to-clock faster than UMC.

---

If the cpu temperature on the outside is 15 o4 more degrees - high chance it will be unstable.
Also, consider starting with 256kb l2 cache.

pshipkov wrote on 2021-08-01, 20:58:

newestest high mark 21.8
Some additional info here.

Cheers

Wait, this was run with only a single stick of 8 MB EDO? What score did you get with 32 MB installed and 64 MB installed?

feipoa wrote on 2021-08-06, 02:18:

Heh, well what do you attribute the 1 fps improvement to with the m918 (21.8 fps) over the UUD board (20.8 fps)? Are you able to provide side-by-side speedsys screenshots?

From your most recent post in the other thread,

m918 at 4x50 MHz, Speedsys:
L1 = 174.9 MB/s ave
L2 = 69.5 MB/s ave
RAM = 69.5 MB/s ave
Quake = 21.8 fps

8433UUD at 4x50 MHz, Speedsys:
L1 = 176.1 MB/s ave
L2 (256k) = 72.9 MB/s
RAM = 53.4 MB/s
Quake =20.8 fps

So I guess the memory read/write speed of the UUD is the limiting factor compared to the m918 and the marginally faster L2 speed on the UUD doesn't help enough because of the size. 1024K might bring them closer together, but I think the m918 would still come out ahead. My tests show around a 0.7 fps gain with 1024K over 256K on the UUD.

pshipkov: Tough to beat the FinALi's memory score. Maybe you can shoot for 4x50 on another M1487/M1489 based board that is not from PC Chips?

feipoa wrote on 2021-08-07, 05:06:

Wait, this was run with only a single stick of 8 MB EDO? What score did you get with 32 MB installed and 64 MB installed?

Yeah, it was unexpected result from a half-assed Saturday morning run.
With 32Mb result is the same. With 64Mb performance drops, as stated in the other thread.

feipoa wrote on 2021-08-07, 05:29:

pshipkov: Tough to beat the FinALi's memory score. Maybe you can shoot for 4x50 on another M1487/M1489 based board that is not from PC Chips?

I don't pay that much attention to these transient metrics, but keep my eyes on the end result - improved applications/games performance.

As I mentioned in the other thread - need to give Abit PB4 a second chance at 180/200, but even if succeed - it will be unlikely to jump higher than M918i.
Didn't know until now about the Gigabyte ALI board.

As for outdoing M918i - i don't think that will be a problem.

pshipkov wrote on 2021-08-07, 07:45:

With 32Mb result is the same. With 64Mb performance drops, as stated in the other thread.

Are you willing to provide a numeric result with 64 MB, with and without SRAM installed ?